Baidu USA LLC

Baidu Online Network Technology (beijing) Co., Ltd.

Yulong Zhang

Tao Wei

Yu Ding

Yueqiang Cheng

Chenfu Bao

Zhenyu Zhong

Qian Feng

Liangzhao Xia

Yunhan Jia

Peng T Li

Peng Li

Minghua Wang

Yongqiang Lu

Shengjian Guo

Tongxin Li

Pei Wang

Longri Zheng

Embodiments of the disclosure disclose a system to prevent data of a client from leaking to untrusted parties in a multiparty computation environment. According to one embodiment, in response to a request received at a gateway (e.g., a non-bypassable gateway) of a server from a user device of a user over a network to process user data by an execution service, the system sanitizes the user data by scanning the user data for malicious code. The system selects a trusted execution environment (TEE) worker from a number of TEE workers and initiates an execution of the execution service by the selected TEE worker. The system receives execution results from the selected TEE worker. The system transmits the execution results to the user device of the user over the network.

TPM-based secure multiparty computing system using a non-bypassable gateway

The present disclosure discloses a method and apparatus for loading a kernel module. A specific embodiment of the method comprises: detecting an instruction for loading a to-be-loaded kernel module into a target kernel; determining, in response to detecting the instruction, whether identifier information is included in the to-be-loaded kernel module, the identifier information being used to represent loading the to-be-loaded kernel module into the target kernel; and loading, in response to determining the identifier information being included in the to-be-loaded kernel module, the to-be-loaded kernel module into the target kernel. This embodiment improves the applicability of the to-be-loaded kernel module.

Method and apparatus for loading kernel module

Embodiments of the disclosure disclose a system to mitigate against adversarial input samples for machine learning (ML)/artificial intelligence (AI) models. According to one embodiment, a system receives a query from a client for a ML service. The system calculates a similarity score for the query based on a number of prior queries received from the client, the similarity score representing a similarity between the received query and the prior queries. The system determines that the query is an adversarial query in response to determining that the similarity score is above a predetermined threshold.

System to mitigate against adversarial samples for ML and AI models

According to one embodiment, a system receives an intermediate result generated by a compiler based on source code, where the intermediate result includes one or more vulnerability indicators indicating one or more lines of the source code being potentially vulnerable. The system performs a grey box fuzzing on a first executable code generated from the intermediate result to generate a first set of seed inputs. The system calculates a vulnerability score for each of the seed inputs of the first set based on the vulnerability indicators for the lines of the source code reachable but has not been explored by the grey box fuzzing. The system selects one of the seed inputs in the first set having a highest vulnerability score. The system performs a concolic execution using the selected seed input as priority, the concolic execution being performed on a second executable code generated from the intermediate result.

Vulnerability driven hybrid test system for application programs

According to one embodiment, in response to a request received at a host agent of a server from a user device of a user over a network to process user data, a data processing system launches a restricted operating environment within the server. The system transmits a token representing the request to a guest agent executed within the restricted operating environment, where an executor associated with the token is launched by the guest agent within the restricted operating environment, where the executor, when executed, is configured to process the user data to generate a processing result without accessing an external component external to the restricted operating environment. The system returns the processing result back to the user device.

Secure multiparty computing framework using a restricted operating environment with a guest agent

According to one embodiment, in response to a request received at a host agent of a server from a user device of a user over a network to process user data, a system transmits a token representing the request to an executor pool having a pool of a number of executors. The system receives by the host agent executable image(s) of an executor from the executor pool, where the executor pool allocated the executor from the pool of executors in response to successfully verifying the token. The system launches a restricted operating environment within the server, including providing the executable image(s) of the executor and the user data to the restricted operating environment. The system executes the executable image(s) of the executor within the restricted operating environment, where the executor, when executed, is to process the user data without accessing an external component external to the restricted operating environment.

Token based secure multiparty computing framework using a restricted operating environment

The present disclosure provides a method, apparatus, device and computer-readable storage medium for guiding symbolic execution. According to embodiments of the present disclosure, it is possible to determine the specific code region of the program, and obtain the program loop output of the program corresponding to the specific code region of the program by using the program inverse analysis method, so that it is possible to obtain the program loop input of the program corresponding to the specific code region by using the program loop predictor according to the program loop output of the program. In this way, the obtained program loop input of the program corresponding to the specific code region may be used to guide the symbolic execution to filter out impossible execution paths and jump out of the program code and reach the specific code region, thereby improving the reliability of the symbolic execution.

Method, device and computer-readable storage medium for guiding symbolic execution

In one embodiment, a system receives a first image captured by a capturing device of an ADV. The system applies an image transformation to the first image to generate a second image. The system applies an object detection model to the first and the second images to generate a first and a second output, respectively. The system calculates a similarity metric between the first and the second output. The system detects the first image as an adversarial sample if a temporal variation of the similarity metric between the first image and a prior image is above a threshold.

Detecting adversarial samples by a vision based perception system

A method, apparatus, and system for using a verification substitute header file in a verification of an application that uses Protocol Buffers (Protobuf) for data communication is disclosed. The operations comprise: obtaining one or more Protobuf message types from a Protobuf definition file; generating the verification substitute header file based on the Protobuf message types; adding one or more verification stubs to the verification substitute header file; including the verification substitute header file in source code of the application; and performing the verification of the application including the verification substitute header file.

Generic verification approach for Protobuf based projects

The present disclosure discloses a method and an apparatus for repairing a kernel vulnerability. A specific implementation of the method comprises: collecting attribute information of a to-be-repaired kernel, the attribute information comprising feature information for representing a code loading mode supported by the to-be-repaired kernel; acquiring a vulnerability repairing code matching the feature information, the vulnerability repairing code being used for invoking a patch application programming interface, the patch application programming interface being used for adapting the vulnerability repairing code to kernel functions of different kernels; loading the vulnerability repairing code and relocating an instruction contained in the vulnerability repairing code; replacing an address relocated by the patch application programming interface in the vulnerability repairing code with an address of the kernel function invoked by the vulnerability repairing code; and executing the vulnerability repairing code with the replaced address to repair the kernel vulnerability.

Method and apparatus for repairing kernel vulnerability

The present disclosure discloses a method and an apparatus for detecting an application. The method comprises: disassembling a binary program of an application running on a target device to generate a to-be-recovered assembly program; selecting, from the to-be-recovered assembly program, a function having function information consistent with function information of a preset reference function in a reference assembly program to obtain at least one candidate function; selecting, among the at least one candidate function, a candidate function having grammatical and/or semantic information consistent with grammatical and/or semantic information of the preset reference function as an object function; selecting, from the object function, a variable having grammatical and/or semantic information consistent with grammatical and/or semantic information of a preset reference variable in the preset reference function as a target variable; and outputting positional information of the object function and the target variable in the to-be-recovered assembly program as a detection result.

Method and apparatus for detecting application

The present disclosure discloses a method, an apparatus and a system for device identification. A specific implementation of the method comprises: receiving a device identification request sent from a terminal device, the device identification request comprising a current user identifier of a current user of the terminal device; acquiring a public key in a preset asymmetric key pair to serve as a first public key; sending the first public key and a randomly-generated first random number to the terminal device; receiving device characteristic information sent from the terminal device, the device characteristic information being generated by the terminal device based on the current user identifier, the first public key, the first random number and a device identifier of the terminal device; and identifying the terminal device based on the current user identifier, the first random number and the device characteristic information.

Method, apparatus and system for device identification

This application discloses at least a kernel repair method and apparatus. An embodiment of the method includes: loading a patch for repairing a to-be-repaired object function in a kernel, the patch being written in a memory-safe language subset of a non-memory-safe language or a memory-safe language; executing the patch using a memory-safe language engine; receiving a call request for calling a repair interface to repair the object function during the executing of the patch, wherein the repair interface is an interface provided by the memory-safe language engine and used for repairing a function; and executing the requested repair interface to repair the object function. As such, the hot patching and security of a kernel can be achieved.

Growing community of inventors

Sunnyvale, CA, United States of America

Yulong Zhang